March 2014 - Boost - lists.boost.org

[asio] Bug: Handlers execute on the wrong strand.
by Greg Barron 04 Apr '14

04 Apr '14

Hi All, I would like to refer you to this bug report: https://svn.boost.org/trac/boost/ticket/9203 I won't paste the full contents of the bug report into this message. The executive summary is that the allocation algorithm for the implementation of asio::strand can and does allocate the implementation of an existing strand to strands that are subsequently allocated. This can have the effect that handlers wrapped in strand (b) will actually be executed on strand (a). I consider this to be a bug but if there is something that I'm misunderstanding about the operation of strands it would be great if you could help me clear it up. If the consensus is that this is a bug, what are the chances for it to be included in the 1.56 release? It is probably too late to be included in the 1.55 release. I've implemented a work around which could only be described as a hack and would be embarrassed to show it to anybody. But it gets me over my hump :). Thanks! -- Greg

5 4

[BGL] Visual Studio 12 finish_edge() failure, general BOOST_TTI_HAS_MEMBER_FUNCTION failure?
by Michael Behrns-Miller 02 Apr '14

02 Apr '14

Hello boost! In the BGL, finish_edge() is a visitor function of DFS, but it is optional. It is made optional via BOOST_TTI_HAS_MEMBER_FUNCTION. Here's how BGL defines it: ------------- BOOST_TTI_HAS_MEMBER_FUNCTION(finish_edge) template <bool IsCallable> struct do_call_finish_edge { template <typename E, typename G, typename Vis> static void call_finish_edge(Vis& vis, const E& e, const G& g) { vis.finish_edge(e, g); } }; template <> struct do_call_finish_edge<false> { template <typename E, typename G, typename Vis> static void call_finish_edge(Vis&, const E&, const G&) {} }; template <typename E, typename G, typename Vis> void call_finish_edge(Vis& vis, const E& e, const G& g) { // Only call if method exists do_call_finish_edge<has_member_function_finish_edge<Vis, void>::value>::call_finish_edge(vis, e, g); } ------------- In all my efforts, the Visual Studio 12 compiler fails to call finish_edge(). I've tried templated versions, direct versions, virtual versions, with and without const params, inline and not inline. In every case, VS says "code will not be hit, no executable code is associated with this line. possible causes: preprocessor directives or compiler/linker optimizations". I'm in debug mode with minimal optimizations, and there's no preprocessor wrapper. It's possible there's a problem with the way I'm calling it, but it happens when I use a simple example from another user. This ticket includes very simple sample code that should fire the finish_edge() function. It does not work for me using VS 12: https://svn.boost.org/trac/boost/ticket/9770 https://svn.boost.org/trac/boost/attachment/ticket/9770/test.cc I have a large project depending on the BGL, which has been great otherwise. I've coded a workaround by using the other non-optional visitor functions. But there are edge cases (sorry, no pun intended! ha) that fail with my workaround. The only possible solution is to get finish_edge() to fire. I'd like to do it the intended way, rather than rewrite BGL code to make finish_edge() non-optional. I'll go try this with a recent version of gcc now... Thanks for any input at all! Michael Behrns-Miller

2 6

Proposal: Vector-like container, which takes O(log(N)) for insert/erase
by Alexander Kuprijanov 01 Apr '14

01 Apr '14

Hi, everybody! I've implemented a container, which has interface similar to vector and takes O(log(N)) for inserting/erasing. Shortly speaking, it has BTrees inside and intrusive leaves, and exploits idea of so-called "rope". I would like to name it alex_kupriianov_array (after myself) and add it to boost. Please tell me who is the coordinator of containers and what is the procedure of adding something to boost. If the community finds it useful, I'll add more features. Among them: 1) non-intrusive version of the container; 2) shallow/partial/lazy copying; if you copy an array and then modify small part of it (for example, beginning), it will keep different beginnings and common end; 3) bookmarks - sort of sticky iterators, which are sticky to a certain element, no matter if there are insertions/deletions before them. Contact me: alexkupri at gmail dot com

6 5

[align] [review] Pre-review comments
by Steven Watanabe 01 Apr '14

01 Apr '14

AMDG I saw the announcement of the review and decided to take a look at the library today. All the following is based on: commit 0d361143c18254efddacfd48775a2776ca7d5e11 Author: Glen Fernandes <glenfe(a)live.com> Date: Thu Mar 27 17:45:12 2014 -0700 Documentation: align: "alignment shall be a fundamental alignment value or an extended alignment value and a power of two." Please define these terms. Also warning: suggest parentheses around ‘and’ within ‘or’ [-Wparentheses] aligned_alloc: I notice that alignment is only required to be a power of two. Is there a reason this is different from align? aligned_allocator: template<class T, size_t Alignment = 1> class aligned_allocator; I'd really like to see the requirements on Alignment written out somewhere, even if we can deduce it from the fact that it calls aligned_alloc. Also, does the standard guarantee that all alignments are powers of 2? Otherwise lcm(2^k, alignof(T)) is not guaranteed to be a power of 2. If it is guaranteed, then lcm(2^k, alignof(T)) == max(2^k, alignof(T)) This is probably worth noting. aligned_allocator notes: "Specifying minimum alignment is generally only suitable for containers such as vector and undesirable with other, node-based, containers." I had to think about this for a few seconds. If you're going to say this, you should probably explain why. aligned_allocator_adaptor: template<class A> aligned_allocator_adaptor(A&& alloc); Not explicit? Should aligned_allocator_adaptor really be convertible from anything? template<class U> aligned_allocator_adaptor(const aligned_allocator_adaptor<U, Alignment>& other); Requires: Allocator shall be constructible from A. Copy/paste error. There is no A here. pointer allocate(size_type n, const_void_pointer hint = 0); Requires: hint is a value obtained by calling allocate(), or else nullptr. Calling allocate on what? Not just this object, but any equivalent aligned_allocator_adaptor, should work. aligned_allocator_adaptor notes "This adaptor class can be used with any C++11 or C++03 allocator including those allocators whose pointer type is a smart pointer." I don't see how this could possibly work with a smart pointer when aligned_allocator_adaptor explicitly uses raw pointers. Your tests pass for we with msvc 9.0-12.0, gcc-mingw-4.6.2, and clang-3.4 on Windows and with gcc 4.7.2 on Linux. aligned_allocator.hpp:100 aligned_free(memory); This is vulnerable to ADL if std::aligned_free is ever created (or my_ns::aligned_free for that matter). (Note that the corresponding aligned_alloc is safe becase std::size_t has no associated namespaces) aligned_allocator.hpp:104 enum { Size = static_cast<std::size_t>(-1) / sizeof(T) }; There's absolutely no reason to use an enum as far as I can see. You can just return this value. aligned_allocator.hpp:139 memory->~U(); I seem to recall msvc warning that memory is unused in cases like this. You may need to supress the warning. detail/align.hpp:28 if (n1 <= space && size <= space - n1) You might add a comment about avoiding overflow, so that no one is tempted to refactor it into an simpler "equivalent" test. test/Jamfile.v2: This is a lot more verbose than it needs to be. Try just: import testing ; run align_test.cpp ; run aligned_alloc_test.cpp ; ... test-suite is obsolete (it just groups all the targets under a single name, which is completely pointless when that's all you have in the Jamfile) and the braces just create a compound statement (like in C++) test/align_test.cpp: You need to test the following: - There is not enough space in the buffer. - In particular, you need to validate the exact boundaries where align starts to return 0. test/aligned_alloc_test.cpp: You need tests for the following conditions: - size == 0 - size < alignment - size > alignment && gcd(size, alignment) != alignment - the allocation fails because size is too large. - Also write to the allocated memory to (hopefully) verify that the data structure used by aligned_alloc is safely outside this region. test/aligned_allocator_adaptor_test.cpp: You're only testing allocate/deallocate. You also need tests for the other members. - rebind - copy constructor/copy assignment operator - construct/destroy - allocate with a hint - allocate with 0 size test/aligned_allocator_test.cpp: same as aligned_allocator_adaptor_test.cpp. test/is_aligned_test.cpp You need tests for: - is_aligned returns false. Your entire test suite would pass with bool is_aligned(...) { return true; } In Christ, Steven Watanabe

3 5

symbolic link missing in modular-boost/boost/numeric
by David Bellot 01 Apr '14

01 Apr '14

Hi, it seems the symbolic link to ublas is missing in "modular-boost"/boost/numeric. I would be very grateful if someone with more admin rights than me could add it and commit it please ? Many thanks. Best regards, David

2 2

[git][iostreams] Forking without an established merge point
by Bjorn Reese 31 Mar '14

31 Mar '14

If I look at the develop branch of Boost.Iostreams then GitHub says that "This branch is 253 commits ahead and 91 commits behind master". It appears that Boost.Iostreams has not established a merge point between master and develop as described in: https://svn.boost.org/trac/boost/wiki/PostCvtMergePoint I am trying to fork Boost.Iostreams to create some patches for it. Normally I would need to work off the develop branch (or a branch originating from develop), but if I do so my pull-requests will include the entire commit history of the component. If I look at the network graph of Iostreams, it appears that everybody else branches off the master branch. How do I proceed?

2 2

[GSOC 2014] Boost.Thread / Work-Stealing-Thread-Pool
by Shuxiong Ye 30 Mar '14

30 Mar '14

Hi boost community, My name is Shuxiong Ye, and I am applying GSOC2014. My proposal is here[1]. There are very few comments on it. Any suggestions or questions will be welcome to improve it. Thanks. [1] https://google-melange.appspot.com/gsoc/proposal/review/student/google/gsoc… Shuxiong Ye

2 3

New numeric data type proposal
by Damian Alberto Vicino 30 Mar '14

30 Mar '14

Hi, I work in a simulation tools' research group and we developed a basic data type for handling time representation that we want to share. The data type is kind of a mix between MultiPresicionFloatingPoint and Rational. We are using this kind of data type in concrete simulator implementations because of a theoretic problem with the formalisms we implement, in which the floating point error can not be bounded. We published a paper in Simutools14 a few days ago explaining how this unbound errors are reached under a specific formalism (DEVS) and why they are not easy to detect. I can share the paper if someone interested in the details, or provide some toy examples to discuss. And we are working in showing same kind of errors in other formalisms. A simplified version of the proposed data type is: Number<N, D, M, E> where in the common case the four are some integer-like type. The number can be interpreted as N/D * M * 2^E where N and D are never modified unless required for operation of compatibility. Compatibility operation is called when an operation between 2 numbers having different N/D happens. It can also be thought as a variable radix MPFP. We choose this implementation because when used as replacement of integer or float, it doesn’t decrease significantly the performance. It only decrease significantly when operating with values that were not representable in any of those (if we simplify N/D). I will love to discuss implementation details, alternative ideas to handle the problem (this one worked for us, but sure there is other options too that may be more generic or performant in generic scenarios). Also, I’m up to code the whole thing, and to share the naive implementation we got in our simulator so far to start the discussion. Best regards, Damian Vicino

3 11

[spirit] How to detect Spirit.Qi expressions?
by Andrey Semashev 30 Mar '14

30 Mar '14

Hi, Is there a way to detect at compile time that a certain type is a Boost.Spirit.Qi parser? That includes template expressions and grammars. Basically, I want to detect if qi::parse(b, e, expr) would be a sensible thing to do. I tried this: boost::is_same< typename proto::domain_of< ExpressionT >::type, spirit::qi::domain > But domain_of returns default_domain for Boost.Spirit.Qi template expressions.

2 2

Getting fully uptodate
by Paul A. Bristow 29 Mar '14

29 Mar '14

Sync SHA-1: 44d48257a7eb39837880afb81ceb2153e86e5a8c * Changed Boost.Log to Boost.Sync in the supported compilers section. Fixes #9388. Current HEAD detached, you are working on (no branch). Do you want to create branch now? Chose "no" [detached HEAD 3a9d142] deletions 37 files changed, 4971 deletions(-) delete mode 100644 include/boost/sync/condition_variables/condition_variable.hpp delete mode 100644 include/boost/sync/condition_variables/cv_status.hpp delete mode 100644 include/boost/sync/condition_variables/notify_all_at_thread_exit.hpp delete mode 100644 include/boost/sync/detail/condition_variables/basic_condition_variable_windows.h pp delete mode 100644 include/boost/sync/detail/condition_variables/condition_variable_posix.hpp delete mode 100644 include/boost/sync/detail/condition_variables/condition_variable_windows.hpp delete mode 100644 include/boost/sync/detail/config.hpp delete mode 100644 include/boost/sync/detail/footer.hpp delete mode 100644 include/boost/sync/detail/header.hpp delete mode 100644 include/boost/sync/detail/interlocked.hpp delete mode 100644 include/boost/sync/detail/link_config.hpp delete mode 100644 include/boost/sync/detail/mutexes/basic_mutex_windows.hpp delete mode 100644 include/boost/sync/detail/mutexes/mutex_posix.hpp delete mode 100644 include/boost/sync/detail/mutexes/mutex_windows.hpp delete mode 100644 include/boost/sync/detail/pthread.hpp delete mode 100644 include/boost/sync/detail/system_error.hpp delete mode 100644 include/boost/sync/detail/throw_exception.hpp delete mode 100644 include/boost/sync/detail/time_traits.hpp delete mode 100644 include/boost/sync/detail/time_units.hpp delete mode 100644 include/boost/sync/detail/tss.hpp delete mode 100644 include/boost/sync/detail/waitable_timer.hpp delete mode 100644 include/boost/sync/detail/weak_linkage.hpp delete mode 100644 include/boost/sync/exceptions/lock_error.hpp delete mode 100644 include/boost/sync/exceptions/resource_error.hpp delete mode 100644 include/boost/sync/exceptions/runtime_exception.hpp delete mode 100644 include/boost/sync/exceptions/wait_error.hpp delete mode 100644 include/boost/sync/locks/lock_guard.hpp delete mode 100644 include/boost/sync/locks/lock_guard_fwd.hpp delete mode 100644 include/boost/sync/locks/lock_options.hpp delete mode 100644 include/boost/sync/locks/shared_lock.hpp delete mode 100644 include/boost/sync/locks/shared_lock_fwd.hpp delete mode 100644 include/boost/sync/locks/unique_lock.hpp delete mode 100644 include/boost/sync/locks/unique_lock_fwd.hpp delete mode 100644 include/boost/sync/locks/upgrade_lock.hpp delete mode 100644 include/boost/sync/locks/upgrade_lock_fwd.hpp delete mode 100644 include/boost/sync/mutexes/mutex.hpp delete mode 100644 include/boost/sync/traits/is_condition_variable_compatible.hpp Success (78 ms @ 29-Mar-14 15:49:53) All appears OK, I:\modular-boost\libs\sync>git status HEAD detached from 44d4825 nothing to commit, working directory clean but the Tortoise GIT folder icon for /libs/sync is still red (Red ! icon means "modified"?) so I have committed to develop and it is still red :-( Fumbling, I have tried a git.exe submodule update --init -- "libs/sync" No submodule mapping found in .gitmodules for path 'libs/sync' git did not exit cleanly (exit code 1) (875 ms @ 29-Mar-14 17:09:10) Suggestions as to what I am misunderstanding/doing wrong to get develop branch to a fully uptodate state? Paul PS I have a similar problem with /tools/regression/build where there appear to be no modifications and nothing to revert, but the build folder icon is red, leaving me uneasy that a red face will follow ;-) I've added the build vcide folder items and committed and finally the icon is green. This doesn't feel something that I should have to do in order to get uptodate... so I must be missing something. It would be quicker to clone from scratch! Paul A. Bristow Prizet Farmhouse Kendal UK LA8 8AB +44 01539 561830 07714330204

2 2